Modernization fails because teams change systems they do not understand.

This platform refuses to let that happen, mechanically. Not through discipline, process or good intentions, but by making the unsafe path impossible to take.

Seven binding principles.

Not values on a wall. A proposed behaviour that violates one of these is wrong, and gets rejected in design review.

Evidence or a question, never an inference

Every fact traces to source code, schema, a runtime observation, or a named human's answer. Nothing is derived from plausibility.

Runtime outranks structure, structure outranks documentation

Old documents describe intent, not behaviour. Where sources conflict, the more direct observation wins and the conflict is recorded.

No change without a way to prove it safe

Code cannot be modified until behaviour-pinning tests exist for what it currently does.

Tasks carry their own context

A task is a self-contained envelope, not a title someone has to go and research before they can start.

Decisions are recorded, not discussed

Migration choices become constraints that bind every downstream task and every agent dispatched afterwards.

The user is always the authority

The platform recommends, explains and blocks. It never decides, and it never overrides on its own initiative.

Language-agnostic by architecture

Language-specific logic lives only in pluggable analyzers. Everything above the model layer is language-neutral.

A meeting room seen through glass in low light

Who it has to serve

An underwriter has to be able to use this.

The hardest questions in a legacy migration are not technical. They are questions about why a business rule exists, and the people who know the answer do not read code. If the platform cannot get an answer out of a claims manager in ninety seconds, the whole evidence chain stops at the semantics layer.

Role	Uses the platform to	Cares most about
Modernization engineer	Pick up tasks, drive agents, review and merge	Task context quality and workspace speed
Technical lead or architect	Choose migration options, sequence work, watch risk	Decisions, dependencies, blast radius
Original system expert	Answer questions about why the legacy behaves as it does	Being asked good questions, briefly
Business or domain owner	Rule on business behaviour questions	Zero jargon and clear stakes
Programme manager	Track progress, blockers and coverage	Blocked work and verification status
Security and compliance	Audit what changed and on whose authority	Provenance and the deployment model

What this is not.

Scope discipline is a feature. Every item below is something the product could plausibly grow into and deliberately will not.

NotA source control hostYour repositories stay where they are

NotA general-purpose development environment competing on editing featuresThe editor was never the defensible part

NotA greenfield development toolEverything here assumes a system that already runs

NotAn autonomous migration serviceThe user decides, always

NotA documentation generatorDocumentation is a by-product of assertions

NotA replacement for expert workshopsIt makes them targeted and captures the output as queryable assertions

Assessments produce documents. This produces executable work.

Static analysis platforms map structure but never observe runtime. Observability platforms see runtime but not intent. Code assistants generate changes without any model of the system. Consulting-led assessments produce slides. Each addresses one slice, and the slices have never been joined.

On naming: Core is the shared evidence store and Hermes is the clarification engine. Both are internal component names, isolated to those two roles, and can be renamed without touching anything else.